Anthropic warns about the increased risks of artificial intelligence

Anthropic, the American artificial intelligence company, has warned that the security risks associated with its models have increased, according to a recently published report.

The risk assessment has been changed from "very low" to "low" regarding dangerous behaviors in high-stakes situations, following incidents observed during security testing. In July, the Claude models gained unauthorized access to the internet and to the computer systems of some organizations.

The report also mentions the existence of an internal model, called "Model 2", which is significantly more performant than Mythos 5, but will not be released publicly. Additionally, Anthropic has observed an increase in the autonomy of the models in research activities, which can have both positive effects and risks.

Despite these challenges, the company will continue the development of artificial intelligence systems, investing in advanced capabilities and security mechanisms.